The threat sensitivity hypothesis predicts that prey species assess and adjust their behavior flexibly in accordance with the magnitude of the threat imposed by a predator. We tested this hypothesis with regard to escape behavior and thanatosis (feigning of death to escape predation) in larvae of the damselfly Ischnura elegans. We manipulated the perceived predation threat of the larvae by changing three factors: lamellae autotomy (an escape strategy where animals sacrifice a body part when grasped by a predator; lamellae present or absent), kairomone type (odors released by predators; control, dragonfly kairomones or fish kairomones), and population of origin (fishpond or fishless pond). We demonstrated that thanatosis increased survival both when confronted with dragonfly and fish predators. We could show, for the first time, costs of past autotomy to be predator-dependent: larvae without lamellae suffered higher predation mortality but only in the presence of a dragonfly predator and not in the presence of a fish predator. This is in accordance with the observed reduced escape speed of larvae after autotomy, which may affect escape probability toward dragonfly predators but not to the very fast fish predators. Unexpectedly, kairomone type did not affect the escape response of the larvae. In accordance with the threat sensitivity hypothesis, after an unsuccessful attack, larvae without lamellae had a higher frequency to enter thanatosis than larvae with lamellae and larvae from the fishpond showed longer thanatosis durations than larvae from the fishless pond. Consistent with the hypothesis, the reaction of the larvae to a simulated attack depended jointly on lamellae status and population. In fishless ponds, larvae with lamellae swam away more frequently than larvae without lamellae; in fishponds both groups almost never swam away and relied mostly upon immobility. Given the obvious benefits of adaptively varying escape responses we hypothesize this threat sensitivity to be widespread. Moreover, we argue that former inconsistencies between studies with regard to escape behavior may have been partly because of such adaptive variation.